Methodology
How we score creators.
Every number in the product traces back to evidence you can see: the videos themselves. Here is exactly how content becomes a score.
Ingest
We index a creator's recent public videos: the footage, captions, stats, and comments. No private data, no platform credentials.
Extract six signals
AI converts each video into structured signals: transcript, on-screen text, comment intent, visual context, language and dialect, and brand-safety markers.
Score components
Signals roll up into creator-level component scores on a 0 to 5 scale: content quality, storytelling, and brand safety.
Match your brief
At search time, your brief is parsed into hard filters and content intent. Brief fit reflects how strongly a creator's actual content matches, never just follower counts.
What each score measures
- Content quality
- Production consistency, clarity of delivery, and how well videos hold attention. Derived from visual analysis and watch-pattern signals across the analyzed sample.
- Storytelling
- Narrative structure: hooks, pacing, and whether product mentions feel natural inside the content or bolted on.
- Brand safety
- Language, claims, and content risk. Specific flags (for example mild language) are always listed on the profile, so you see why a score is what it is.
- Engagement signals
- Comment rate and save rate relative to views, computed directly from the analyzed videos and shown against benchmarks. High saves signal reference-worthy content; rich comments signal purchase intent.
Reading the 0 to 5 scale
- 4.5 - 5.0
- Exceptional. Consistently strong across the analyzed sample.
- 4.0 - 4.4
- Strong. Reliable quality with minor variance.
- 3.5 - 3.9
- Solid. Good content with notable inconsistency or flags worth reviewing.
- Below 3.5
- Review carefully. Flags or quality variance need a human look before briefing.
Detecting suspicious followers
Follower counts are the easiest number to fake, so we treat them as a claim to verify, not a fact. Each follower base is checked against behavior patterns that separate genuine accounts from bots and engagement farms. The signals we weigh:
- Missing or generic profile pictures
- Extreme following-to-follower ratios
- Account age and history
- Posting activity and cadence
- Empty or templated bios
- Sudden, unnatural follower growth
Who is actually behind the follower count
A follower count is a mix of fans, passers-by, and noise. Every profile splits it into five groups, so you can price the reach you are actually buying:
- Genuine audience
- Active, real accounts that watch, comment, and save. The people your campaign actually reaches.
- Amplifiers
- Real followers with an audience of their own. When they engage, the content travels beyond the creator's page.
- Passive followers
- Real accounts that follow thousands of profiles. They count toward reach on paper but rarely see a post.
- Inflated accounts
- Mass-following profiles with follow-for-follow or purchased-growth patterns. Reach you should discount.
- Likely bots
- Automated accounts: empty profiles, zero history, machine-generated handles. No value to a campaign.
Honest limits, beta methodology
- Scores are AI-assessed estimates from a sample of recent videos, not audits of a creator's full history. They update as new videos are indexed.
- We analyze public content only. Some platform metrics (for example Instagram saves) are not public; where shown, they are estimated from observable signals and labeled with benchmarks.
- Audience breakdowns are statistical estimates from observable account signals, not a census of every follower.
- Scores predict content fit, not campaign outcomes. A 4.7 is a strong starting shortlist, not a guaranteed result.
- The methodology is in beta and will be versioned. When the scoring changes, profiles will say so.